Hi Guys, this is my first attempt at making something public, so my apologies if this seems completely amateur and unprofessional, because it absolutely is!
So what I've done here is Dreambooth train 2 different models on SD1.4, one on images of Gelflings from the Dark Crystal, and another model on Skeksis. I then merged both models together with equal weights, and this is the result. (training done via NMKD Stable Diffusion on an RTX 3090 for 4000 steps for both models)
class words are "Gelfling" and "Skeksis" - I've found that it tends to really favor the bird beaks for the Skeksis, so if you're finding beaks on everything, de-emphasize by .3 or .4. On the inverse of that, I've found I need to emphasize gelflings to about 1.2 to really get good gelfling examples. Word of warning, it has no clue what is male and what is female for both classes, so don't be upset by cross dressing gelflings!
Here are Skeksis Samples, some with Gelflings as well:
and here are the Gelflings (bonus points if you can figure out the celebrity likenesses!)
Hope you enjoy, I'm impressed with how well this training turned out. I look forward to seeing Gelflings in the wild!